A Cognitive Radio Tracking System for Indoor Environments

Advances in wireless communication have enabled mobility of personal computing services equipped with sensing and computing capabilities. This has motivated the development of location-based services (LBS) that are implemented on top of existing communication infrastructures to cater tochanging user contexts. To enable and support the delivery of LBS, accurate, reliable, and realtime user location information is needed. This thesis introduces a cognitive dynamic system for tracking the position of mobile users using received signal strength (RSS) in Wireless Local Area Networks (WLAN).The main challenge in WLAN positioning is the unpredictable nature of the RSS-positionrelationship. Existing system rely on a set of training samples collected at a set of anchor pointswith known positions in the environment to characterize this relationship. The first contributionof this thesis is the use of nonparametric kernel density estimation for minimum mean square errorpositioning using the RSS training data. This formulation enables the rigorous study of state-spacefiltering in the context of WLAN positioning. The outcome is the Nonparametric Information (NI)filter, a novel recursive position estimator that incorporates both RSS measurements and a dynamicmodel of pedestrian motion during estimation. In contrast to traditional Kalman filteringapproaches, the NI filter does not require the explicit knowledge of RSS-position relationship andis therefore well-suited for the WLAN positioning problem.The use of the dynamic motion model by the NI filter leads to the design of a cognitive dynamictracking system. This design harnesses the benefits of feedback and position predictions from thefilter to guide the selection of anchor points and radio sensors used during estimation.Experimental results using real measurement from an office environment demonstrate theeffectiveness of proactive determination of sensing and estimation parameters in mitigatingdifficulties that arise due to the unpredictable nature of the indoor radio environment. Inparticular, the results indicate that the proposed cognitive design achieves an improvement of3.19m (56\%) in positioning error relative to memoryless positioning alone.